Does Schema Markup Really Improve Rankings and AI Citations?

Schema markup has been a standard technical SEO recommendation for years.

If you have a product, add Product schema. If you publish articles, use Article schema. If you represent a company or local business, add Organization or LocalBusiness structured data.

More recently, another claim has become common:

“If you want to appear in ChatGPT, Google AI Mode, Perplexity and other AI engines, you need schema markup.”

But does the evidence actually support that?

After looking at research from Ahrefs, Search Atlas, Semrush and others, alongside Google's own guidance, the most defensible conclusion is:

Schema is useful infrastructure for describing information to machines, but strong evidence that schema alone directly increases organic rankings or AI citations is currently lacking.

That distinction matters.

Schema being useful is not the same as schema being a ranking or citation factor.

What schema markup actually does

Schema markup is structured data that describes the meaning of information on a webpage in a machine-readable format.

JSON-LD is the most commonly used implementation.

A product page, for example, can explicitly describe:

• product name,

• brand,

• price,

• availability,

• rating.

Schema.org also includes types such as Article, Organization, LocalBusiness, Recipe, Event, JobPosting and Breadcrumb.

Google says structured data can help its systems understand page content and make pages eligible for richer search experiences.

However, structured data by itself is not a generic Google ranking factor.

Does schema improve Google rankings?

Google's answer has remained remarkably consistent:

Not directly.

Google's John Mueller reiterated in 2025 that structured data does not make a site rank better. It is primarily used for supported search features.

That means adding perfect JSON-LD does not give your page an automatic ranking bonus over a competitor without schema.

This does not make schema useless.

Product, Recipe, Event, JobPosting and other supported structured-data types can still be important for specific Google search experiences and rich results.

The value is therefore often closer to:

better representation in search results, rather than a direct ranking boost.

What about AI search engines?

This is where the debate becomes more interesting.

A common GEO/AEO assumption is that structured data should make it easier for language models and AI retrieval systems to understand content, therefore increasing the probability of being cited.

The theory makes sense.

The evidence is much less straightforward.

The correlation initially looked extremely strong

Ahrefs initially analyzed roughly six million URLs.

Pages cited by AI were almost three times more likely to contain JSON-LD than pages that were not cited.

That sounds like powerful evidence for schema.

But it has a major problem:

Correlation does not prove causation.

Websites that maintain structured data well are also more likely to have:

• stronger technical SEO,

• better content,

• more backlinks,

• better site architecture,

• stronger authority,

• professional SEO teams.

Schema may simply be traveling alongside the factors that actually cause greater visibility.

Ahrefs therefore ran a second study designed to isolate schema itself.

The 1,885-page Ahrefs test

Ahrefs identified 1,885 pages that added JSON-LD between August 2025 and March 2026.

They matched them against roughly 4,000 control pages with similar citation behavior and compared citation changes before and after schema implementation.

The platforms measured were:

• Google AI Overviews,

• Google AI Mode,

• ChatGPT.

The measured effects were approximately:

PlatformEffectGoogle AI Overviews-4.6%Google AI Mode+2.4%ChatGPT+2.2%

The small positive differences for AI Mode and ChatGPT were statistically indistinguishable from zero.

In other words:

Adding schema did not produce a meaningful increase in AI citations.

There is an important limitation.

These pages were already receiving substantial AI citations.

So the research is good evidence that:

Adding schema to an already visible page does not suddenly produce a large citation boost.

It does not definitively answer whether structured data might help a brand-new or previously invisible page get discovered or interpreted for the first time.

Search Atlas found a similar pattern

Search Atlas compared domain-level schema coverage with visibility across OpenAI, Gemini and Perplexity.

Its analysis found that increasing schema coverage did not reliably correspond with increasing LLM visibility.

The researchers concluded that schema alone is unlikely to be a major driver of LLM citation behavior.

That result broadly supports the Ahrefs findings.

Why Semrush appears to show something different

Semrush published research comparing characteristics found in cited and non-cited content.

The strongest positive associations included:

• clarity and summarization: +32.83%

• E-E-A-T signals: +30.64%

• Q&A formatting: +25.45%

• section structure: +22.91%

• structured-data elements: +21.60%

At first glance, that appears to contradict Ahrefs.

It does not necessarily.

Semrush is effectively asking:

“What characteristics are more common among cited content?”

Ahrefs is asking:

“If we add schema to an existing page, do citations increase?”

Those are different questions.

Well-maintained, authoritative content may be more likely both to contain structured data and to be cited by AI.

That still does not prove that structured data caused the citation.

FAQ schema did not outperform in another dataset

Research reported by Search Engine Journal found that several commonly promoted AI optimization tactics performed weakly.

Pages with FAQ schema averaged around 3.6 citations, compared with 4.2 citations for pages without it.

LLMs.txt also showed negligible impact.

That does not prove FAQ schema is harmful.

It does, however, weaken the claim that simply adding FAQ structured data reliably increases ChatGPT citations.

Does Google require schema for AI Overviews or AI Mode?

Google's official guidance is very clear:

No.

There are no additional technical requirements for appearing in AI Overviews or AI Mode beyond the usual Search requirements.

Google specifically says you do not need:

• a special AI schema,

• special AI markup,

• an AI text file,

• llms.txt,

• a separate machine-readable format.

Google recommends continuing to use structured data when appropriate, especially for regular Search and rich results, but explicitly states that structured data is not required for generative AI Search.

Instead, Google continues to emphasize:

• crawlability,

• indexability,

• internal linking,

• useful content,

• text accessibility,

• good page experience,

• consistency between structured data and visible content.

Does ChatGPT even see JSON-LD?

An experiment published by Dejan SEO examined whether structured data reached the model when a page was fetched for an OpenAI browsing request.

The experiment found no evidence that the JSON-LD schema reached the model; the retrieved context appeared to contain ordinary page text instead.

Hartzer raises a similar technical argument.

JSON-LD normally exists inside a <script> element.

Many HTML-to-text conversion systems remove script elements before passing page content into a model.

That makes it entirely plausible that schema could disappear during some real-time retrieval flows.

But this should not be exaggerated.

OpenAI, Anthropic and Perplexity do not publicly document every stage of their indexing and retrieval pipelines.

Structured data could potentially be used during indexing, entity extraction, knowledge-graph construction or another preprocessing layer even if it does not enter the model's context during an individual fetch.

What about studies that support schema?

There is also evidence on the other side.

Whitehat SEO cites a Growth Marshal analysis of more than 50,000 articles.

It reports citation rates of:

• attribute-rich schema: 61.7%

• no schema: 59.8%

• generic schema: 41.6%

The Plus Addons similarly argues that Ahrefs may underestimate schema's usefulness for pages starting with little or no AI visibility because the Ahrefs treatment group already consisted of heavily cited URLs.

These findings are worth considering.

But they should not be treated as equivalent to a matched controlled experiment.

Whenever a study claims schema causes higher AI visibility, ask:

• Was there a control group?

• Was the same page measured before and after implementation?

• Were authority and content quality controlled?

• Were other SEO changes happening at the same time?

• Has the experiment been independently reproduced?

Those questions separate useful evidence from attractive correlation.

The interesting exception: LocalBusiness and ChatGPT

A controlled LocalBusiness experiment discussed by CMSWire produced a particularly interesting result.

Across 29 domains and 36 locations, advanced LocalBusiness structured data did not meaningfully improve rankings across traditional Google, Bing, Yahoo or Maps measurements.

ChatGPT was different.

The test reported an improvement of roughly 3.3 positions and around 10 percentage points of Share of AI Voice.

One experiment is not enough to establish a universal rule.

But it suggests something important:

Schema effects may depend heavily on the platform, query and entity type.

For local recommendation queries, clearly structured business entities may be considerably more useful than generic Article schema on an informational page.

Should you remove schema?

No.

That would be the wrong conclusion.

Schema still has several legitimate uses.

Rich results

Product, Recipe, Event, JobPosting and other supported schema types continue to power specific Google search experiences.

Entity definition

Organization, Person, Product and LocalBusiness markup can explicitly describe:

• who an organization is,

• who an author is,

• what a product is,

• which URLs and profiles refer to the same entity.

That is useful data modeling even without a ranking bonus.

Data consistency

Implementing structured data properly encourages consistency across:

• authors,

• canonical URLs,

• prices,

• availability,

• dates,

• entity identifiers.

Future machine consumers

Google is not the only consumer of structured information.

Search engines, commerce platforms, browser agents and future AI systems may process structured data in different ways.

Maintaining accurate structured data can therefore remain a sensible infrastructure decision.

The mistake is treating schema as an AI ranking hack

The wrong strategy is:

“The more schema types I add, the more likely ChatGPT is to cite me.”

There is no convincing evidence for that.

Unnecessary structured data can instead create:

• maintenance overhead,

• conflicting information,

• mismatches between visible content and JSON-LD,

• developer work that could have been spent fixing larger SEO problems.

Google explicitly recommends that structured data accurately match the visible content of the page.

What should you prioritize for AI citations?

Interestingly, the strongest signals across several studies are not exotic technical tricks.

Semrush found stronger associations for clarity, E-E-A-T, Q&A formatting and section structure than for structured-data elements.

Google continues to emphasize ordinary Search fundamentals for generative AI experiences as well.

A more sensible priority order is therefore:

1. Answer the user's question better than competing pages.

2. Make important answers explicit in visible text.

3. Structure the page with meaningful headings.

4. Add original evidence, examples, expertise and sources.

5. Make sure the page can be crawled and indexed.

6. Fix broken links, unnecessary redirects and 404s.

7. Build strong internal links between related entities and topics.

8. Add appropriate structured data.

The order matters.

Schema cannot compensate for weak content or broken technical SEO.

Where No404 fits into this

In AI search, a machine first has to successfully reach the information before it can understand or cite it.

If a crawler or retrieval system encounters:

• a 404,

• an outdated URL,

• a broken internal link,

• an unnecessary redirect chain,

• an incorrect destination,

perfect schema will not solve the underlying access problem.

That is why systems such as No404, which detect missing URLs, analyze broken links and help route old requests toward relevant live destinations, should be thought of as a technical accessibility layer beneath structured data.

First make sure machines can reach the correct resource.

Then make the visible content clear.

Then use structured data to describe what that resource represents.

Conclusion: Does schema matter?

For most websites:

Yes, use it where it has a legitimate purpose.

But do not implement it under the assumption that:

“Adding schema will automatically increase rankings or AI citations.”

The strongest current evidence does not support that statement.

Ahrefs found no meaningful citation increase after schema implementation across ChatGPT or Google AI Mode. Search Atlas found that greater schema coverage did not reliably predict LLM visibility. Google explicitly states that structured data is not required for generative AI Search.

At the same time, structured data remains useful for:

• rich results,

• commerce data,

• local entities,

• entity definition,

• interoperability between systems.

The better question is therefore not:

“Should we use schema?”

It is:

“Is there a real consumer for this structured data, and can we measure the benefit?”

If the answer is yes, use it.

Just do not mistake infrastructure for a ranking strategy.

As of 2026, that is the conclusion best supported by the evidence:

Schema is infrastructure. It is not a magic AI visibility switch.